Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77319.24020064223,
-      "required_pages_per_second_upper_estimate": 0.005349124442434929
+      "required_token_positions_per_second": 77271.05334122997,
+      "required_pages_per_second_upper_estimate": 0.005345790763691871
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9664.90502508028,
-      "required_pages_per_second_upper_estimate": 0.004046493907833771
+      "required_token_positions_per_second": 9658.881667653746,
+      "required_pages_per_second_upper_estimate": 0.004043972053861288
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19329.81005016056,
-      "required_pages_per_second_upper_estimate": 0.004025889672077454
+      "required_token_positions_per_second": 19317.763335307493,
+      "required_pages_per_second_upper_estimate": 0.004023380659066795
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28994.715075240838,
-      "required_pages_per_second_upper_estimate": 0.0017381888963942215
+      "required_token_positions_per_second": 28976.64500296124,
+      "required_pages_per_second_upper_estimate": 0.0017371056231524624
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28994.715075240838,
-      "required_pages_per_second_upper_estimate": 0.002254211562826597
+      "required_token_positions_per_second": 28976.64500296124,
+      "required_pages_per_second_upper_estimate": 0.002252806694188706
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28994.715075240838,
-      "required_pages_per_second_upper_estimate": 0.37191307288568437
+      "required_token_positions_per_second": 28976.64500296124,
+      "required_pages_per_second_upper_estimate": 0.3716812894006136
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.2464871291015669,
-    "producer_p95_seconds": 0.1909218159999,
-    "gpu_update_seconds": 0.9535530847002244,
-    "producer_batches_per_second": 4.057006966834128,
-    "gpu_batches_per_second": 1.0487093126172178,
-    "producer_headroom": 3.8685715078749836,
+    "producer_batch_seconds": 0.19570271662501426,
+    "producer_p95_seconds": 0.19453867399988667,
+    "gpu_update_seconds": 0.9541477281850449,
+    "producer_batches_per_second": 5.109791101756134,
+    "gpu_batches_per_second": 1.048055736507568,
+    "producer_headroom": 4.875495571240772,
     "sustainable": true,
-    "buffer_seconds": 60.073844336114135,
+    "buffer_seconds": 60.11130687565783,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386451.3235817268,
-    "upload_bytes_per_second": 5830306.941499609,
+    "generation_bytes_per_second": 386709.7715296907,
+    "upload_bytes_per_second": 5939729.325420699,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.92345791424611,
+    "estimated_full_shards_per_hour": 82.97891482751885,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -406,6 +406,12 @@
       "bytes": 2455057,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2129529189996902,
+      "bytes": 11955081,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -50,3 +50,4 @@
 {"step": 490000, "loss": 0.6224940419197083, "lr": 0.0003, "grad_norm": 0.16324780468432967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71719430326, "tokens_per_second": 151494.72346230017, "data_wait_s": 7.556000127806328e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 490100, "loss": 0.5918187499046326, "lr": 0.0003, "grad_norm": 0.1435987313990175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71733993967, "tokens_per_second": 138910.31397707752, "data_wait_s": 1.3097999726596754e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 490200, "loss": 0.7364625930786133, "lr": 0.0003, "grad_norm": 0.2280283186482364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71748771623, "tokens_per_second": 154282.14708244952, "data_wait_s": 8.301999514515046e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 490300, "loss": 0.5763944983482361, "lr": 0.0003, "grad_norm": 0.18574894482687884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71763351892, "tokens_per_second": 152421.45844116813, "data_wait_s": 2.0497999685176183e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:63d433fc99be4d2f82c71d50e155c68242f10611244cab94f9b9123db1f034bc
-size 111084
+oid sha256:95adce0f963166414276de653e27882669691421177ddcfdf0edde2fb9fc5801
+size 113202